User guide

Getting started

• Getting started
• Supported systems

Core concepts

• Tensor types
• Scalar tensors
• Matrix and Vector
• Advanced
• Per-thread matrix and vector operations
• Per-thread linear algebra
• Tensors
• Compound types
• BufferView
• qd.static
• Sub-functions
• Parallelization

Integration

• Numpy and Torch interop
• Shared Metal command queue (PyTorch MPS)

Autodiff

• Automatic differentiation

SIMT primitives

• GPU execution model
• Atomics
• Block primitives
• Grid primitives
• Math
• Subgroup primitives
• Register-resident tiles: Tile16x16 and Tile32x32

Algorithms

• Algorithms

Performance

• Performance
• Fastcache
• Graph
• Streams
• Performance Dispatch
• qd.init options

Testing

• Advanced: Unit testing
• Kernel code coverage

Reference

• Python features in kernel scope
• Python backend
• Debug mode
• Quirks
• Troubleshooting

Advanced

• Optimization passes

Contributing

• Advanced: Contributing to quadrants

Internal

• Advanced: Building the CUDA graph conditional fatbin